Skip to content

Commit caf2035

Browse files
authored
Merge pull request #2158 from lidge-jun/codex/windows-verification-2270
docs(devlog): record the local Windows verification behind the 2.27.0 promotion
2 parents 70e8bab + 94ecc29 commit caf2035

1 file changed

Lines changed: 55 additions & 0 deletions

File tree

Lines changed: 55 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,55 @@
1+
# 200 - local Windows verification before promoting 2.27.0
2+
3+
Run on the maintainer Windows machine at `dev` = `70e8bab42` (v2.27.0),
4+
Bun 1.3.14, Windows 11. This is the operator-side check that 190 asked for
5+
before promotion, recorded with its evidence rather than summarized after.
6+
7+
## Results
8+
9+
| Gate | Result |
10+
|---|---|
11+
| `bun install` | clean, no changes across 105 installs |
12+
| `bun run typecheck` | exit 0 |
13+
| `bun run test` | 8028 pass, 1 fail, then a Bun runtime panic |
14+
15+
The suite took 2021s against a ~210s idle baseline, so the machine was heavily
16+
contended. That matters for reading the result: slowness alone did not produce
17+
the failure below, which is why it was re-run in isolation.
18+
19+
## The one failure, and why it is not this release's
20+
21+
`tests/codex-app-server-processes.test.ts` - "a defaulted read is memoized, and
22+
invalidation is what clears it" (#1046).
23+
24+
It reproduces standalone in six seconds, so it is not a contention artifact.
25+
It is also not caused by `main...dev`: checking out **`origin/main`'s copy of
26+
both `src/codex/app-server-processes.ts` and the test file** and re-running
27+
produces the identical single failure. Same defect, older code.
28+
29+
The mechanism is environmental. This machine runs live Codex app-servers, so
30+
the probe finds real processes and returns `unknown`, which carries the
31+
deliberate 250ms `CATALOG_STATE_UNKNOWN_TTL_MS` window instead of the 5s one.
32+
The second call lands outside that window, recomputes, and returns a
33+
structurally equal but distinct object - hence "serializes to the same string".
34+
A CI runner has no Codex app-server running, reaches `not_running`, and gets
35+
the full 5s TTL, so the case passes there and fails only on a developer box
36+
that is actually using Codex.
37+
38+
Worth stating plainly: the test asserts object identity through a cache whose
39+
TTL depends on what the machine happens to be running. That is a real test
40+
defect, not a product defect, and it belongs to #2152's family rather than to
41+
this release.
42+
43+
## The trailing panic
44+
45+
The run ended with `panic: Internal assertion failure` inside Bun itself, after
46+
the last test file reported. This is the shard-2 Bun runtime panic already
47+
tracked in [#2152](https://github.com/lidge-jun/opencodex/issues/2152) - a
48+
runtime crash, not a test result.
49+
50+
## Verdict
51+
52+
Promotion criterion as stated in 190 and confirmed at the audit gate: zero
53+
failures **outside** the tracked #2152 set. Met. Typecheck is green, the suite
54+
is green except for one pre-existing environment-dependent case proven against
55+
`origin/main`, and the panic is a known tracked crash.

0 commit comments

Comments
 (0)